n8n pricing in 2026: the real cost, not the sticker price

Every article about n8n pricing answers the wrong question. They list the tiers, note that the Community edition is free to self-host, and stop — as though the subscription were the cost. It is not. On the stack costed below, the n8n Cloud subscription is 5.5% of what n8n actually costs to run for a year. The other 94.5% is people.

This matters because the sticker price is what people budget against. A company decides to adopt n8n, approves a subscription in the low tens of euros, and books it as a small line in the software budget. Four months later somebody notices that a developer has spent most of a quarter on it. Nothing went wrong — the plan was simply costed against the wrong number, and the wrong number was off by a factor of eighteen.

This article prices the whole thing: license, hosting, and the expensive line nobody puts in a pricing table. Every figure has arithmetic you can check, and every variable input is labeled as an assumption so you can substitute your own. There is no proprietary benchmark here and no survey — just three assumptions and a lot of multiplication, which means you can disagree with an input and recompute rather than having to take our word for a total.

Last updated: 7 September 2026.

The assumptions in this article

Three numbers here are inherently local to you. They are stated once, used consistently, and flagged wherever they appear:

• Loaded hourly rate: $65. What an hour of your own technical person actually costs, including employer taxes, benefits, tooling, and idle time — not their salary divided by 2,080. Substitute yours; it moves the totals more than anything else on this page.
• Server cost: $40 a month for a 4 vCPU / 8 GB machine, plus $15 for managed Postgres. Cloud providers differ by a factor of three for the same specification.
• n8n Cloud tier: $55 a month for the tier covering our example volume. n8n’s entry tier is around €20-25 a month and the vendor changes its packaging regularly, so check the current page. Everything else on this page is arithmetic, not estimation.

We take no vendor commission and resell no n8n licenses, so nothing below is written to move a subscription.

The three cost lines

If you take one thing from this page: the lines are ordered by how much attention they get and reverse-ordered by how much they cost.

Line 1: the license and the n8n Cloud subscription

The short version, and it is genuinely simple.

Self-hosted Community edition: free to license. It runs under n8n’s Sustainable Use License, which permits internal business use and restricts reselling it as a service. If you are automating your own company’s work, you are inside the license. If you plan to resell n8n workflows to clients as a product, read the license properly first.

n8n Cloud: paid, starting in the low tens of euros per month. The entry tier sits around €20-25. Tiers scale on execution allowance and the number of concurrently active workflows, which is the packaging detail that catches people: you can be well inside your execution allowance and still need a higher tier because you have too many workflows switched on.

The billing model is the part worth understanding, and it is n8n’s genuine structural advantage: n8n bills per workflow execution, not per step. One run of a workflow is one billable unit whether it contains three nodes or fifty. Make bills per operation and Zapier bills per task, meaning both charge for every step inside every run. That comparison is the whole subject of n8n vs Make vs Zapier, and if you have not chosen a platform yet, read that first — this article assumes the decision is made. If you are not yet sure what n8n does, start with what is n8n.

The practical consequence for cost: on n8n you never have to choose between a debuggable workflow and a cheap one. Validation nodes, logging steps, and explicit error branches are free. On per-step billing they are a line item, which is how workflows end up with no error handling and a bill that still grows.

Two counting details worth knowing before you size a tier. A workflow that fails halfway and gets retried is generally another execution, so a flaky integration inflates your usage as well as your support load — budget a retry allowance rather than your ideal-world run count. And a workflow that loops internally over a batch of two hundred records is still one execution, which is the single largest lever you have over the subscription line: batching a per-record trigger into a scheduled run that processes the whole batch can cut the execution count by an order of magnitude without changing what the automation does.

Put a number on it for the example stack below: $55 a month, or $660 a year.

Line 2: hosting, if you self-host

This is where “free” quietly stops being free. Two sub-lines: the infrastructure, which is cheap and predictable, and the operational hours, which are neither.

The infrastructure

Nothing surprising, and if this were the whole cost of self-hosting then everybody should self-host: $840 a year against $660 for n8n Cloud is close enough that the control would be worth the difference.

Two notes on the specification. Running Postgres on the same box saves the $15 but makes backups your problem in a way most people discover during a restore. And you want more RAM than feels necessary — n8n holds workflow data in memory during execution, and the failure mode of an undersized box is a workflow that dies partway through a batch, which is the worst kind of failure to debug.

The operational hours

Self-hosted total: $840 + $2,860 = $3,700 a year, or about $310 a month.

That second case deserves scepticism. “Our ops team will absorb it” is true in companies with an ops team and a runbook culture, and false in companies where one senior developer is quietly the ops team. If the 44 hours land on one person who has other commitments, you have not saved money — you have moved the cost somewhere it does not get measured, and created a single point of failure who cannot take a holiday during an upgrade window.

Line 3: the expensive one — engineer hours

Here is the line that decides the budget. It exists identically whether you are on n8n Cloud or self-hosted, which is why the Cloud-versus-self-hosted argument is a distraction from the actual money.

Building the workflows

Per production workflow crossing two or three systems:

The happy path is five of twenty-four hours. That ratio is the single most useful number in this article, because it explains why an n8n demo takes an afternoon and a workflow you can leave running unattended takes three days. Everyone who has been burned by automation was burned in the other nineteen hours.

For the four-workflow stack: 4 × 24 = 96 hours, plus about 20 hours of shared setup done once — environments, a credential store, a reusable error-notification subworkflow, logging, alerting. Total: 116 hours, or $7,540 at $65.

Keeping them alive

Integrations change on the vendor’s schedule, not yours.

$3,705 a year to maintain four workflows. That is 5.6 times the n8n Cloud subscription, and roughly the same size as an entire self-hosted server bill including its ops labor. If you budget one line beyond the subscription, budget this one.

Worked example: the same stack, costed three ways

A concrete four-workflow stack for a company of about 40 people.

Note the two right-hand totals: 1,800 billable units on n8n, about 17,300 on per-step billing. Nine and a half times the meter for identical work. That is the platform comparison argument in one row.

Now the three routes, year one:

What that table actually says

Read it honestly, including the parts that do not flatter us.

Done-for-you is the most expensive column in year one. $17,320 against $11,905 — about $5,400 more. Anyone selling you agency work as the cheapest option is not showing you this table.

The subscription is a rounding error in every column. $660 out of $11,905 is 5.5%. Per execution: the n8n Cloud subscription costs 3.1 cents, while the true all-in cost is 55 cents on the Cloud DIY route, 69 cents self-hosted, and 80 cents done-for-you. If you are choosing an n8n tier by comparing cents, you are working on the wrong line.

Self-hosting loses by $3,040 at this scale. Not because servers are expensive, but because 44 hours are.

The DIY column rests entirely on one assumption: that 173 hours is the real number. Work out the crossover. Done-for-you at $17,320 equals the Cloud DIY route once the in-house hours reach 256 — $17,320 minus the $660 subscription, divided by $65. That is a 48% overrun on the estimate. A team building its first production automations overrunning by 48% on exception handling is not a pessimistic scenario; it is the median one. The DIY column is cheaper if your hour estimate holds and more expensive if it does not, and you will not know which until month four.

There is a fourth column, and it is the interesting one. Pay for the build, maintain it yourself: $7,600 build + $660 subscription + 57 hours of your own maintenance at $3,705 = $11,965. That is within $60 of building it yourself. Our four-workflow build price is, almost exactly, what 116 in-house hours cost at $65 an hour — so the build is free in the only sense that matters, and what the retainer buys is the 57 hours, not the 116.

That equivalence only holds because our prices sit at the market floor. At market rates — $6,000-15,000 per workflow build and $3,000-8,000 a month for a retainer — the same four workflows and twelve months come to $60,000 to $156,000, before an audit fee of $2,500-8,000. Against that, doing it yourself is obviously correct, and the arithmetic is not close.

When doing it yourself is genuinely cheaper

The honest cases, and they are common. Do not hire anyone — including us — if:

• You have one or two simple workflows, not four. A single trigger and two actions is an afternoon. The 24-hour figure above is for a workflow crossing multiple systems with real exception paths; do not apply it to something that has none.
• Someone in-house already knows n8n. The learning curve is most of the risk in the DIY column. Remove it and the 173-hour estimate becomes credible, which makes DIY the cheapest route by a clear margin.
• The hours are genuinely spare. If a technically curious person has slack capacity and wants the skill, the $65 loaded rate overstates your real cost, because you are paying it either way.
• The process is still changing. Automating a moving target produces rework you pay for twice. Build the throwaway version yourself, learn the process, and pay someone once it has stopped moving.
• The workflow does not clear the payback bar. If the manual process costs under roughly $300 a month, no route on this page pays back before the process changes. That test comes before any of this arithmetic — it is covered in how much AI automation costs.

Hiring earns its fee in the opposite conditions: three or more systems, real exceptions, an audited approval step, or — most often — a company where nobody has 173 hours that are actually free.

n8n pricing vs Zapier: when the platform bill starts to matter

One caveat to everything above. The platform line stays a rounding error only while volume stays moderate.

Our example runs 1,800 executions and about 17,300 steps a month. On per-step billing that is well into the upper paid tiers rather than the entry ones. Assume — and check the current tiers, because all three vendors change them — that the difference is $500 a month. That is $6,000 a year, which is no longer a rounding error: it is half the entire Cloud DIY total and more than the annual maintenance line.

So the rule has two halves. Below roughly 5,000 billable steps a month, the platform choice barely registers against labor and you should pick on whichever tool your team can maintain. Above about 20,000, the billing model becomes a real budget line and n8n’s per-execution pricing is the reason to move. The multiplication that tells you which half you are in — runs × steps — is the whole of the platform comparison.

What we charge, in public

For the third column, our prices are published rather than quoted:

• Process audit: $490. Market rate $2,500-8,000. Credited in full against a build. It produces the volume and step counts this article’s arithmetic needs, plus an honest verdict on whether the process should be automated at all.
• Workflow build: from $1,900 per production workflow. Market rate $6,000-15,000. Fixed price, source and credentials in your accounts, documentation included.
• Retainer: $690 a month. Market rate $3,000-8,000. This is the line that covers the 57 annual hours of integration drift and silent failures — and it is only worth buying once something is actually running.

There is no free pilot. A pilot that costs us nothing produces a demo, and the nineteen hours that separate a demo from a workflow you can trust are exactly the hours a free pilot skips.

If you want that priced against your own stack rather than our example, the n8n agency page has the build scope and the automation service page has the build-and-maintain engagement. Or run the arithmetic yourself first — the ROI calculator takes your volumes and hourly rate and tells you which column of the table above you belong in. If the answer is the DIY column, that is a good outcome, and you should not pay anyone.